Three Days of
Arabic Linguistics

Introduction. This is a study of how prevoicing and aspiration affect the perception of /d-t/ in Khuzestani Arabic(KhA), Hawrami (Haw), and Balochi (Bal). Depending on the used VOT categories, two-way laryngeal systems aredivided into three types: aspiration (short lag vs long lag), true voicing (prevoicing vs short lag), and over-specified(prevoicing vs long lag). We examined two over-specified languages, Khuzestani Arabic (Bahrani & Kulikov, 2023)and Hawrami (Kulikov & Bahrani, in press), and one true voicing language, Balochi (Kulikov & Bahrani, in press).We show that prevoicing plays no role in this perceptual distinction. Although speakers of these languages consistentlyproduce phonologically voiced stops with prevoicing, the results reported here show that this feature is redundantperceptually as listeners instead rely only on the duration of aspiration. To learn more about the perception ofprevoicing in these languages, especially in over-specified languages, the following questions were answered:1) Do listeners rely on prevoicing to classify a stimulus as /d/?2) Where is the perceptual /d/-/t/ VOT boundary?3) What is the effect of increasing the intensity of VOTs on listeners’ perception?4) What are the possible theoretical implications of our findings?Methods. Fifteen native speakers per language, between 20-45 years old, participated in this experiment in-person inIran. Stimuli were made by manipulating natural speech in Praat. VOT intensity was left unchanged or multiplied by4. Value 4 was selected after observing no difference between the original intensity and doubling it in a pilot perceptionstudy. An identification task was run on PCIbex by using the Visual Analogue Scale task instead of the two-alternativeforced choice as it can better capture the gradient effects of prevoicing in responses. Participants were presented witha scale on the screen with ‘d’ word on the right end and ‘t’ word on the left end and heard tokens along the /d/-/t/ VOTcontinuum (Figure 1). They were told to move the circle in the middle closer to the left end or the right end dependingon how similar each stimulus sounds to either word. Data was analyzed by linear regression mixed effects modelswith main effects of VOT, intensity, and their interaction. The response data was empirical-logit transformed.Conclusion. In our test languages the perceptual boundary was bigger than zero in both intensity conditions, althoughthe increase in VOT intensity moved the boundary closer to zero (Figure 2 for KhA). Furthermore, the percentage of/d/ responses to negative VOTs was almost 100% with no gradiency in listeners’ responses. The redundancy ofprevoicing in perception supports the abstract underlying representation of voicing contrast. In addition, it poses anissue for Laryngeal Realism which uses the word-initial VOT categories as the underlying representation of voicingcontrast: ‘does the failure of prevoicing to influence perception in all three languages undermine representing /d/ as[voice] phonologically?’. Using prevoicing in production can be a way of marking sociophonetic differences amongdifferent dialects/languages (e.g. Herd, 2020).

Forced alignment is a staple tool for research in phonetics and phonology, offering preciseword- and phone-level alignment of audio signals with pre-written transcriptions. Transcriptions,although time-consuming to produce, ensure high levels of accuracy. Even when requiring hand-correction, mediocre alignments can greatly speed up data processing for phonetic analysis. Onlyone of the most used forced aligners, MAUS (Kisler et al., 2017), has any Arabic support at all,but only provides word-level alignment and transcription for Modern Standard Arabic (MSA).This study presents first steps toward word- and phone-level forced alignment for dialectal Arabic.The forced aligner is trained through the Montreal Forced Aligner (MFA) framework,which allows users to create their own models (McAuliffe et al., 2017). MFA requires four inputsto align data: audios, transcripts, pronunciation dictionary, and acoustic model. The audio andtranscripts originate from the Massive Arabic Speech Corpus (MASC), a 1200-hour corpus ofsubtitled Arabic-language YouTube videos (Al-Feytani et al., 2023). Preliminary results come froma subset of twenty-four hours of MSA audio and eight hours of Egyptian (EG) audio, classified as“clean” (versus “noisy”) by the authors of MASC.The pronunciation dictionary was generated by modifying a large MSA pronunciationdictionary (Doherty, 2016). Each of the over 857,000 Arabic headwords in the dictionary (oftenmorphologically complex) is accompanied by one or more phonetic transcriptions. This dictionarywas modified in Python to accommodate Egyptian phones in a manner that allows for futurecustomization for any dialect. The preliminary focus is on consonants, so the modifications allowusers to specify which phone or phones each Arabic letter corresponds to. For example, for anMSA dataset, users can specify that ج corresponds to /dʒ/, for an Egyptian dataset it maycorrespond to /g/, and for a Levantine dataset it may correspond to /ʒ/ (Youssef, 2021). For adataset with multiple dialects, all three phones may be included. The script then generates newdictionary entries. For example, the word جميل “beautiful,” could have any combination of thepronunciations [dʒamiːl, ʒamiːl, gamiːl] etc., depending on what phones ج was specified for.The pronunciation dictionary was generated with multiple phones for five targetconsonants that differ between MSA and EG: ث (MSA [θ], EG [t]), ج (MSA [dʒ], EG [g]), ذ (MSA[ð], EG [d]), ظ (MSA [ðˁ], EG [dˁ], and ق (MSA [q], EG [ʔ])(Watson, 2002; Haddad, 2023). Otherpronunciations in the dataset were ignored for simplicity in the early stages of developing thealigner. Once the dictionary had been generated, an acoustic model was trained using MFA’s trainfunction and then applied to the data. To analyze training performance, counts of each transcriptionwere tabulated (Table 1).Table 1. Counts (and percentage) of MSA and Egyptian Arabic phones for the five target letters.Contrary to expectations, only for ق did the Egyptian data have a larger number of theexpected phone than the MSA data. These results suggest that the alignment training was not verysuccessful. However, performance can be improved in future models through retraining the modelwith some corrected reference alignments, increasing the dataset size, and performing additionalcleaning/normalization prior to training. In spite of the unsuccessful model demonstrated here, theability to create custom, dialect-specific pronunciation dictionaries overcomes one major hurdlein the creation of accurate forced alignment.

Despite extensive work on Arabic triliteral roots, the internal organization of quadriliteral roots (QRs)remains underanalyzed. Greenberg (1950) mentions them only briefly, essentially noting that they obey thesame cooccurrence regularities as triconsonantal roots. Also, in both early Arab grammars and modernscholarship, QRs are viewed as extensions of bi- or triconsonantal roots formed through reduplication oraugmentation and are often consigned to onomatopoeic periphery (Bohas, 1997; El Zarka, 2005;Kuryłowicz, 1973). For example, reduplicated /zlzl/ ‘shake repeatedly’ is traced back to the biradical verb/zl/ ‘slip’, while an augmented root such as /šmxr/ ‘be proud’ may result from affixation, radical insertion,or the fusion of two triliteral roots. Many studies also note that QRs appear in old and recent loanwords(e.g., /sndn/ ‘anvil’ (Persian); /tlfn/ ‘telephone’ (French)) (Ferguson, 1959; Heath, 1989; Mahadin, 1996).Beyond these descriptive claims, what the literature lacks is a systematic typology of QRs and, crucially, aquantitative account of their structure and the principles (if any) that shape it.We address this gap by examining relative sonority as a potential organizing principle in the structure ofArabic QRs. We propose three types of QRs: augmented, reduplicated, and primitive. Augmented QRsinvolve radical insertion as in /šmx-r/ ‘be proud’ from /šmx/ ‘be high’ (Wright, 1933:49), a process thatcontinues to be productive (as in /rqm-n/ ‘digitize’ from /rqm/ ‘digit’). Reduplicated QRs typically featurerepetition of primitive biliteral roots (e.g., Moroccan Arabic /žržr/ ‘drag’ from /žr/ ‘pull’) but may alsoderive from etymological triliteral roots (e.g., /hmhm/ ‘to murmur’ from /hms/ ‘whisper’), often expressingiterative, plural, or mimetic meanings (e.g., /zqzq/ ‘bird tweet’). Primitive QRs, by contrast, aremorphologically non-derived and semantically independent.To examine the phonotactics of Arabic QRs, we compiled a corpus of 1,575 primitive quadriconsonantalroots (C1C2C3C4) from Arabic dictionaries and modern usage sources. We calculated the sonority transitionswithin each bigram, C1-C2, C2-C3, and C3-C4 and found that QRs adhere to a distinct sonority frame. Arepeated-measures ANOVA over our data revealed a significant effect of position on sonority (F(2, 2650)= 6.42, p = .003), suggesting that while sonority tends to increase from C1 to C2 and from C3 to C4, itdecreases from C2 to C3. We argue that this rise-fall-rise pattern reflects an active sonority-based constraintthat extends beyond the surface CaCCaC prosodic template associated with Arabic QRs (see /zalzal/,/šamrax/, /raqman/), and further propose that C1C2 and C3C4 clusters form distinct phonological domains,or sonority ‘shells’, reflecting the likely historical evolution of QRs through reduplication of some minimallexical unit consisting of a sonority wave/contour with a sonority onset and a sonority peak/head.We show that, in addition to its diachronic reality, this sonority frame has always been active in the mentalgrammar of Arabic speakers: it explains why both old and recent loan QRs tend to conform to this sonoritypattern. A comparison of native primitive and loan QRs in our corpus reveals a striking parallel, suggestingthat loan QRs that make it into the lexicon exhibit the same frame (e.g., Moroccan Arabic /srkl/ ‘walkaround’ from French cercle ‘circle’; Gulf Arabic /bnšr/ ‘flat tire’ from English puncture). A non-parametricFriedman test comparing sonority distances across consonant positions indicated a significant effect ofposition on sonority distance in borrowed QRs (χ²(2) = 142.88, p < .001). Interestingly, while both primitiveand borrowed QRs displayed a rise-fall-rise sonority pattern across positions, the effect was markedlystronger in borrowed roots (χ²(2) = 142.88, p <.001, W=.27) than in primitive roots (χ²(2) = 11.57, p =.003,W=.009), indicating a stronger positional sonority-based constraint in synchronic grammar. Additionally,data shows that borrowed QRs undergo phonological adaptation so that they fit in the expected sonorityframe. For instance, in sankar ‘to scan’, /snkr/ is extracted from scanner but underwent a /k/-/n/ metathesis;and in darzan ‘dozen’, /drzn/ was obtained from through insertion of a sonorant /r/ in C2 (not in C3).We discuss the broader implications of our findings for the synchronic nature and genesis of Arabicquadriconsonantal roots and for the broader models of phonological templates, abstract phonologicalconstraints on lexical units, and loanword adaptation in modern Arabic.

Heritage speakers (HS) are individuals who grow up in households where a minority language is spoken and acquirethe majority community language during early childhood (Montrul, 2016). Research on a variety of languages,including Hindi, Mandarin, Spanish, and Korean, demonstrates that heritage speakers often outperform secondlanguage (L2) learners in perceiving phonemic contrasts, a benefit attributed to early exposure to heritage languageinput (Tees & Werker, 1984; Au et al., 2002; Knightly et al., 2003; Oh et al., 2003; Godson, 2004). In Arabic,emphatic consonants are a defining feature of the phonological system. These sounds are produced with a secondaryconstriction in the pharyngeal or velar region, which distinguishes them from their plain counterparts (Watson,2002). Research on vowel context and consonant position has shown that these factors affect the perception of thesecontrasts, as coarticulatory effects and emphasis spread can modulate their perceptual salience (Jongman et al.,2011; Al-Masri & Jongman, 2004; Hayes-Harb & Durham, 2016). Recognizing these factors, the present studyinvestigates how vowel context and consonant position influence the perception of emphatic–plain contrasts inArabic heritage speakers of Levantine descent compared to English-speaking L2 learners. We hypothesized thatheritage speakers would outperform L2 learners in accuracy due to early exposure to the heritage language.Participants included eighteen Arabic heritage speakers (of Jordanian, Syrian, or Palestinian descent) and eighteenL2 learners (American), who were undergraduates enrolled in intermediate or advanced university Arabic courses.The stimuli consisted of monosyllabic Arabic words sampled from university-level textbooks, containing emphatic(/dˤ/, /tˤ/, /sˤ/, /ðˤ/) and plain (/d/, /t/, /s/, /ð/) consonants in both word-initial and word-final positions, across threeshort vowel contexts (/æ/, /u/, /i/). A female native speaker of Levantine Arabic recorded all the words. During theauditory forced-choice identification task, participants were instructed to respond as quickly and accurately aspossible. In each trial, they heard a single word and responded to the prompt “Which word did you hear?” Forexample, they distinguished between [sˤæb] ‘he poured’ (PST.3MSG) and [sæb] ‘he pulled’ (PST.3MSG).Data were analyzed using logistic regression, with Group (HS vs. L2), Consonant type (emphatic vs. plain), Voweltype (/æ/, /u/, /i/), and Word position (initial vs. final) as predictors, along with their interactions. Heritage speakersdemonstrated significantly higher identification accuracy (M = 74.7%, SD = 9.2) than L2 learners (M = 59.5%, SD= 8.7; Group (L2 – HS): β = -1.2041, SE = 0.299, z = -4.023, p < .001), indicating an advantage conferred by earlyexposure. Consonant type had a marginally significant effect (β = 0.7152, SE = 0.373, z = 1.916, p = 0.055), withbetter performance for plain (70.0%) than emphatic (63.2%) consonants. A significant interaction between group andword position was observed (β = 0.5734, SE = 0.292, z = 1.965, p = 0.049): heritage speakers showed similaraccuracy in initial (75.5%) and final (73.8%) positions, while L2 learners’ accuracy declined from initial (65%) tofinal (53.9%) positions. There was also a significant interaction between consonant type and vowel context (/u/ –/æ/, β = -1.6426, SE = 0.432, z = -3.800, p < .001), with plain consonants identified most accurately in the /æ/context and emphatic consonants more accurately in the /u/ context. An interaction between vowel context and wordposition indicated that accuracy differences between initial and final positions were more pronounced for /æ/ thanfor /u/. Reaction times were examined using linear regression with the same predictors. While the main group effectwas not significant (β = -920.6, SE = 1111, t = -0.83, p = 0.407), a significant interaction emerged between groupand consonant type (β = 5001.8, SE = 1571, t = 3.18, p = 0.001): heritage speakers showed a smaller reaction timedifference between emphatic and plain consonants, whereas L2 learners had a larger disparity. Furthermore,significant three-way interactions among group, consonant type, and vowel context were found ((i – æ): β = -5153.4,SE = 1924, t = -2.68, p = 0.007; (u – æ): β = -5183.0, SE = 1924, t = -2.69, p = 0.007), indicating that heritagespeakers’ reaction times were stable across vowel contexts, while L2 learners’ varied according to both consonantand vowel.These findings highlight the perceptual advantages of early exposure for heritage speakers, while also suggestingthat continuous input is crucial for undergoing native-like processing. Neither group achieved the accuracy levelreported by native speakers in Jongman et al. (2011). The absence of a word position effect among heritage speakerssuggests the use of more generalized perceptual strategies than those seen in other heritage language contexts (cf.Oh et al., 2003, for Korean). This study presents new evidence of how linguistic experience influences Arabicspeech perception and offers implications for second language acquisition, heritage phonology, and curriculumdevelopment.

Background. The Interface Hypothesis explains why certain constructions are more susceptible to crosslinguistic influence andprotracted instability in advanced L2 learning (Sorace, 2005; Sorace & Filiaci, 2006; Sorace, 2011; White, 2011). Constructions inwhich syntax interfaces with other internal modules (syntax-semantics interface), are differentiated from external interfaces, in whichsyntax interfaces with other “higher” modules (syntax-pragmatics/discourse interface). The focus on the L2 acquisition of interfaceproperties has recently extended to explore crosslinguistic influence and the learnability conditions that modulate the protractedinstability of these properties (Sorace, 2005; Sorace & Serratrice, 2009; Valenzuela, 2006).Interface Case in Question. The cases in question are the definite and bare plurals in MSA and English. Work on the syntax-semantics interface of definiteness in MSA and English has shown clear differences in how definite and bare nominals are interpretedin preverbal and postverbal positions (Fehri, 2004). Three interpretative possibilities are discussed: generic, specific, and existential.For genericity or kind reference, English uses three noun phrases in the preverbal position, one of which is bare plurals. Moreimportantly, a generic reading is not available for definite plurals as illustrated in (1). In MSA for generic reference to types, definitesingular and plural nouns are interchangeably used in the preverbal position (Fehri 2004) as illustrated in (2) and (3). Interestingly, inMSA bare plurals are grammatical in the preverbal position, but they are assigned an existential reading as illustrated in (4). This caseis commonly construed to be an instance of internal interface.Research Questions. (1) To what extent do three English-speaking learner groups of Arabic exhibits crosslinguistic transfer in theproduction of generic definite plurals? This question will be addressed by comparing production data of three groups of L2 Arabic;And (2) does input exposure as a learnability condition modulate protracted instability of generic definite plurals? This question willbe addressed by comparing production data for advanced learners who studied Arabic in an at-home and abroad. A native speakergroup was used as a control group in both questions.Data and Results. Data were collected using a prompted sentence completion task that asked the participants to read each sentence tounderstand the meaning established and provide the single missing word (spoken and written) with the help of a picture. This was inaddition to a prompted oral narrative task that elicited generic reading and the specific reading at the sentence level. Both tasks arecommonly held to provide spontaneous data in L2 studies. Accuracy rates for the suppliance of the definite article for the first questionfor the two tasks are provided in the tables below. A mixed design ANOVA was conducted on the group results. It returned asignificant main effect for group: F(3,47) = 46.039, p = .000, ηp2 = .746. Also, there was a significant main effect for the readingcondition: F(1,47) = 121.083, p = .000, ηp2 = .720, as well as a significant interaction between group and the reading condition:F(3,47) = 14.189, p = .0000, ηp2 = .475. They overall suggest a considerable crosslinguistic influence from English in the genericreading through providing bare plurals that persisted until advanced learning. This effect resulted in protracted instability of definiteplurals. Only the advanced-high group participants were able to overcome this instability and attain a high rate.This pattern was also found in the results of the oral narrative task. The low-advanced group clearly fluctuated between definite andbare plurals in the generic condition. The high-advanced participants steadily opted for definite plurals and did not show fluctuationbetween definite and bare plurals. Accuracy rates for the suppliance of the definite article for the second research question areprovided in the tables below.For the advanced-AH group, their performance overall was far from being target-like. The average score of target definite plurals wassurprisingly at chance (49.66%; SD: 0.31), and the average score of non-target bare plurals was slightly higher (50.33%; SD: 0.31)with no difference between both averages in a paired-samples t-test: t(9) = 0.9692, p = 0.3608. This pattern of fluctuation betweenbare and definite plurals showed that the advanced-at home participants were exhibiting protracted instability as a manifestation ofpersistent L1 effects. In comparison, the advanced-study abroad participants demonstrated an entirely different pattern. Theydemonstrated a considerable degree of stability, as the average score of their definite plurals was high (93.28%; SD: 0.14), and theaverage score of their bare plurals strikingly dropped: (6.72%; SD: 0.14), with a significant difference between both averages: t(9) = -7.4703, p = 0.0002. The same pattern was found in the oral narrative task.Discussion and Conclusion. The fluctuation patterns that the low-advanced group exhibited in the generic reading condition is takenas a manifestation of the persistent effects of L1 English. It is also consistent with the predictions of the original proposal of theInterface Hypothesis. The integration of semantic knowledge (generic vs. specific) with syntactic knowledge (bare vs. definite plurals)remained far from optimal and gave rise to protracted instability. This instability was caused by competition with the bare plurals intheir L1 English. Ionin and Montrul (2010) and Montrul and Ionin (2012) concluded that the complexities of mapping definite andbare plural nouns to their meanings (whether specific or generic) is one of the reasons this grammatical property is more vulnerable topersistent L1 effects in sentence interpretation even at higher levels of proficiency. The results of the beginning and low-advancedgroups are consistent with the tenets of the Bottleneck Hypothesis by Slabakova (2019), which seeks to explain what is hard and whatis easy to acquire in a second language. It proposes that it is functional morphology (such as definiteness) that is the bottleneck of L2acquisition. The results of the study abroad group supports the role of input exposure in stabilizing this particular case of interface. Itis possible that this group utilized the non-occurrence of generic bare plurals in the preverbal position as a cue in the input to map ageneric reading only to definite plurals. The question of which properties are harder to acquire than others will be further discussed.

“Oh, my beloved,” “chum,” “oh woman,” “my girls’ father,” “oh sheikha,” “my soul,” “oh my unfortunateoutcome”! The art of address terms among lovers and spouses is a profoundly social practice shaped bysocietal perceptions. Performative interactions in Egyptian TV social dramas provide a rare public windowinto these intimate linguistic practices during private moments, reflecting societal perceptions of theircreators, who both mirror and reshape their communities’ linguistic practices in everyday interactions. Thissociopragmatic study provides an original analysis of the nuances of Egyptian address terms from a stylisticperspective, focusing on how TV social dramas depict these terms among couples from intersecting socialbackgrounds (sociolinguistics) and their discursive social goals (pragmatics), comparing them with existingstereotypes about gender. The study demonstrates how 588 tokens of nominal address terms by 24 couplesin six Egyptian TV social dramas (2016-2024) construct significant styles associated with specific socialgroups. The data captures 144 interactions (scenes), totalling 192 minutes, with six interactions lasting eightminutes per couple, where each spouse has a balanced amount of speaking time, across 95 episodes, writtenby 18 authors (9 men and 9 women). The data maintains a balanced representation of couples who areprotagonists with leading roles, belonging to two gender groups (husbands and wives), two age groups(young, 20-39, and middle-aged, 40-59), and three social classes (upper, middle, and working).The concept of ‘style’ was first introduced in sociolinguistics by Labov (1973), then expanded byCoupland (1980, 2007), along with many linguists (e.g., Mejdell, 2006), to identify the distinctive ways ofspeaking associated with social groups by the frequent co-occurrence of a linguistic form with a specificgroup more than others, serving social functions. I argue that address terms are often a stylistic choiceconnecting lexical meanings to particular social groups and discursive goals, and their social use on screenis not arbitrary but governed by social rules. Terms of address have not been explicitly or deeply exploredfrom a stylistic perspective in the Arab world and beyond, primarily due to the emphasis on their lexicalmeanings, either with reference to power and solidarity (e.g., Brown and Gilman, 1960), power andpoliteness (e.g., Brown and Levinson, 1978), sociolinguistics including social variables (e.g., Qin, 2008),or pragmatics including their general or specific functions (e.g., Mühleisen, 2011). Even sociopragmaticstudies on address terms integrating social variables and overall or specific functions demonstrate theseaspects separately without explicitly or thoroughly examining terms of address as a stylistic choice (e.g.,Al-Balqa’, 2021; Parkinson, 1985). This work takes a step further by going beyond binary approaches andcomprehensively combining previously isolated frameworks under the lens of style in the performativelandscape, adding to our socio-pragmatic understanding of the art of address terms in intimate relationshipsin Egypt.The study reveals that love and kinship terms are perceived as the most significant stylistic forms ofaddress among couples in Egyptian Arabic. Social identities, such as gender, age, and social class, interactto influence love and kinship terms, with no single identity being the most significant. The Egyptian TVseries construct these intersecting social identities among couples through these terms. Love terms ofaddress, like ‘/ya ḥabebti/’ (my beloved), distinguish young upper-class men when addressing their wives.In contrast, kinship terms of address, such as ‘/ya xōya/’ (chum, literally brother), characterize middle-agedworking-class women when addressing their husbands. Both young upper-class men and middle-agedworking-class women instrumentalize these styles to achieve discursive social goals and manage theirmarriage relationships rather than simply expressing love or referring to actual familial relationships. Suchportrayals both negotiate and reinforce broader claims about gender. For example, young upper-class men’sstyle of love terms of address negotiates the common perception about men as emotionally unlettered (e.g.,Cameron, 2008). Meanwhile, the style of middle-aged working-class women in using kinship terms, alongwith their primary social goal of blaming, criticizing, and complaining, reinforces the common perceptionthat women often express negative emotions and complaints (e.g., Talbot, 2010). The study encouragesfurther research on how styles of address terms are constructed and reconstructed, both in spontaneous andperformed speech, across various relationships in the Arab world and beyond.

This paper employs Critical Discourse Analysis to examine the linguistic evolutionwithin the Egyptian Muslim Brotherhood's political discourse, specifically its shift fromadvocating for the Khilāfah to promoting a Dawlah Madaniyyah (civil state). By analyzing keyArabic texts and statements from figures like Ḥasan al-Bannā, Sayyid Quṭb, ʿAbd al-Qādir ʿUda,and Muḥammad Mahdī ʿĀkif, the study investigates how changes in word choice, modality, andintertextual references reflect broader ideological transformations.Using Fairclough’s three-dimensional CDA model, the analysis highlights the discursivestrategies employed by Brotherhood leaders to construct notions of legitimacy, governance, andreligious authority. The findings indicate that while the term Khilāfah became less frequent, itscore concepts—unity, shūrā (consultation), and Sharīʿah—were recontextualized within themore adaptable framework of the Dawlah Madaniyyah. This linguistic change, I argue,represents not a break from previous ideology but rather a rephrasing of Islamic political idealsto suit the demands of modern political discourse.This study contributes to Arabic discourse studies by illustrating how political-religiousmovements navigate modernity through linguistic adaptation, offering insights into theideological functions of Arabic political terminology in post-Islamist settings.

Vocatives are expressions used to attract the hearer’s attention to the proposition of an utterance(Lambrecht 1996). This study examines the Tunisian Arabic particle ha:, a lesser-knownvocative counterpart to ya:, as in ya: Sonia / ha: Sonia. While ya: has been analyzed in previouswork (e.g., Al-Bataineh 2020; Haddad 2020; Shormani & Qarabesh 2018), ha: has not beeninvestigated in any capacity. I argue that although both particles introduce vocative phrases, onlyha: serves an additional pragmatic function: it functions as an attitudinal marker, guiding thehearer to recognize the speaker’s emotional stance toward the addressee.Corpus evidence shows that ha: is more constrained than ya: in frequency and distribution. Its use islimited to utterances in which the speaker expresses a negative or critical attitude toward the addressee.For example, in (1), a YouTube commentator uses ha: to mock an actress who changed her Arabic name,Jalila, to the Latinate name, Julia, conveying disapproval and sarcasm.(1) hhhh ha: ʒali :la rak men tu:nis. esm-ik ʒalilaaaaahhhh. VOC Jalila EMPH From Tunisia name-your Jalilaaaaa‘hhhh Jalila, you are from Tunisia, [and] your [real] name is Jalilaaaaa ’In contrast, when the speaker expresses a positive or neutral (balanced) attitude toward the addressee,only ya: is felicitous, whereas ha: results in infelicity, as shown in (2) and (3). In (2), the speakerexpresses support for the president by urging him not to reverse his decision to suspend the parliament.Here, ya: appropriately conveys respect and solidarity, while ha: would convey an unintendedinterpretation of criticism or mockery. Similarly, in (3), a blogger asks other participants for advice onhow to add friends on a social media platform. The context provides no evidence of a positive or negativestance toward the addressees. Using ha: in this context would therefore introduce an unnecessaryevaluative meaning such as blaming others for being unhelpful and could lead to miscommunication.(2) la: ruʒu:ʕ ya:/#ha: siya:dit r-raʔi:sNEG going back VOC Mr. the-president‘There is no going back, Mr. President.’(3)Ya:/#ha: ʒma:ʕa ʃniyya l-ħall mtƐ:ʕha l-ħka:ya haðiVOC folks what the-solution POS the-matter this‘Folks, what is the solution to this matter?’Building on research on attitudinal particles, this study adopts Relevance Theory (Sperber & Wilson1986/1995) to argue that the Tunisian Arabic particle ha: semantically encodes a procedural instructiondirecting the hearer to recognize the speaker’s emotional or attitudinal stance toward the addressee. In doingso, ha: serves as a guarantee of relevance, explicitly prompting the hearer to seek additional contextualassumptions to arrive at the intended interpretation. Thus, while ya: can occur in contexts similar to (1),only ha: explicitly instructs the hearer to interpret the utterance as conveying an emotional attitude.

Wodak & Meyer (2001) state that “language is not powerful on its own – it gains power by the usepowerful people make of it.” In line with this view, the present study investigates the language ofmanipulation employed by the most influential presidential candidates in their TV interviews during the2012 Egyptian presidential election. At that time, Egypt had never experienced a democratic regime, andthe 2012 election was deemed the first free, competitive presidential election in its history. This led manycandidates to employ manipulative strategies in an attempt to influence public opinion and gain electoralsupport.This study examines the intersection of pragmatics and critical discourse analysis (CDA) to uncovermanipulative strategies used by Egyptian presidential candidates following the 2011 revolution. ThroughTV interviews from the 2012 election, five leading candidates—Mohamed Morsi, Ahmed Shafiq, AmrMoussa, Hamdeen Sabahi, and Abdelmonem Aboulfotoh—were analyzed using a pragma-criticalframework that integrates conversational implicature, presupposition, and speech acts with Van Dijk’ssocio-cognitive model and Aristotle’s rhetorical appeals (ethos, pathos, logos).By merging micro-level pragmatic insights with macro-level ideological critique, the analysis reveals howcandidates strategically used language to construct credible personas, trigger emotions, and achievemanipulative discourse aimed at getting elected. The study also explores how positive self-representationand negative othering were used to manipulate public perception.This interdisciplinary approach demonstrats the value of integrating pragmatics and CDA in politicaldiscourse research. It highlights how applied linguistics can engage with real-world challenges. Finally, itpositions political discourse as a space where language evolves, ideologies are shaped, and people’svalues and beliefs can be influenced.

This paper investigates how two binominal constructions in standard Arabic—bare noun-noun compounds andconstruct state nominals—pattern with and diverge from their well-studied Hebrew counterparts, and what thesesimilarities and differences imply for theories of word formation, and whether or not word formation is a distinctive,sequential module within the generative process. Building on Borer’s (1988) proposal that word formation is amodule that operates at multiple levels of generative grammar, the presented Arabic data, especially in the domain ofidiomatic and calqued expressions, calls for an enriched account that incorporates insights from ConstructionMorphology (Booij 2010).Borer’s analysis of Hebrew distinguishes between bare compounds (e.g. beyt xolim [house sicks] ‘hospital’) andconstruct state nominals (e.g. ca’if ha-yalda [scarf the-girl] ‘the girl’s scarf’) on several dimensions: definitenessmarking, the locus of primary stress, the availability of internal modification, and the behavior of number features onthe head and complement nouns. In her account, bare compounds are formed lexically and exhibit semantic opacityconsistent with the Lexical Integrity Hypothesis, while construct state nominals are syntactically formed andphonologically unified, with features such as plurality and definiteness percolating via a process she termsSecondary Percolation. These patterns motivate the view that word formation is not confined to a single pre- or post-syntactic component, but can apply at multiple points in the derivation.Arabic features an analogous pair of binominal constructions. As in Hebrew, both can mark plurality anddefiniteness on the head, and both show the genitive marking of the complement noun. However, a closer look at thesemantic behavior of Arabic constructions complicates a straightforward adoption of Borer’s architecture. Arabicexhibits (i) compounds with relatively transparent, often modern calque-like meanings (e.g. ālat zamān [machinetime] ‘time machine’), and (ii) construct state nominals with non-compositional, often lexicalized or metaphoricalmeanings (e.g. ’eid al-mīlad [festival the-birth] ‘Christmas’; rijl al-qanṭur [centaur the-foot] ‘(the constellation)Alpha Centauri’). This distribution contrasts with the strongly lexical-idiomatic character of Hebrew compoundsthat underpins Borer’s division of labor between lexical and syntactic word formation.A survey of Arabic binominal expressions shows that idiomatic or non-compositional meanings in Arabic can besystematically associated with particular constructional schema. Non-compositional construct state nominals tend todenote named categories and specific entities (epithets, species, toponyms, astronomical objects), while manycompounds function as calques for technical tools, artifacts, or abstract concepts modeled on European sourcelanguages. These patterns suggest that, in Arabic, idiomaticity is not tightly tied to a single structural configuration(compound vs. construct) but to higher-level constructional schemas that cut across the traditional lexical/syntacticdivide.I therefore propose that Borer’s insights about feature percolation, opacity, and word-formation strata can bemaintained, but that the interpretive component must be enriched by constructional schemas of the sort posited inConstruction Morphology. For Arabic, we can posit at least two such schemas: (i) a “calque-compound” schema thatlicenses N–N sequences denoting instantiations of tools, implements, or conceptual types, typically with anindefinite genitive complement; and (ii) a “non-compositional construct” schema that licenses definite construct statenominals interpreted as metaphorical exemplars of a category possessed or modified by an attribute. Genitive caseand (in)definiteness then conspire with these schemas to yield the observed contrast between specific, lexicallyanchored referents and more general, instantiated concepts.The broader theoretical implication is that a purely generative, structure-only account of word formation isinsufficient once we consider cross-linguistic variation in idiomaticity and calquing strategies. Arabic shows that thesame structural template (e.g. construct state nominal) can host both compositional and non-compositionalmeanings, depending on its constructional schema. Semitic word formation can be better understood through amulti-level word-formation module with a construction-based semantics. Future directions for this proposal includea more rigorous corpus-based investigation of colloquial varieties, as well as further research into the full range ofArabic calquing strategies and how a more complex calquing system might inform the outlined schemata.

In Arabic NLP, most work on morphological processing has focused on Modern Standard Arabic (MSA) and a few major dialects such as Egyptian, Levantine, and Gulf Arabic. Researchers initially developed MSA-based morphological analyzers, including early systems such as the Buckwalter Arabic Morphological Analyzer (BAMA; Buckwalter, 2002), the Standard Arabic Morphological Analyzer (SAMA; Maamouri et al. 2010), and later Farasa (Abdelali et al. 2016).Subsequent tools expanded coverage to include dialectal morphology, especially for Egyptian, Levantine, and Gulf Arabic. MADAMIRA (Pasha et al,. 2014) was among the first analyzers to incorporate dialectal variants, while the more recent CAMeL Tools (Obeid et al., 2020) integrates advanced analyzers such as CALIMA-Star. Despite these developments, coverage of many dialects remains limited.Moroccan Arabic (MA) is one such dialect that remains underrepresented in tool development and evaluation. MA exhibits distinctive and complex morphological phenomena, one of which is diminutive formation. Diminutives are frequent in everyday speech and play an important role in highlighting the morphological richness of MA. Their complexity arises from irregular phonological processes—such as reduplication, syllable insertion, gemination, and degemination—that the base form undergoes(Elmdari,1999;Boudlal,2001;Lahrouchi & Ridouane, 2016). These irregularities pose serious challenges for existing Arabic NLP tools.Despite their frequency and linguistic significance, diminutive forms have received little to no computational attention. This lack of coverage reveals a significant gap in how current morphological analyzers process Moroccan Arabic. Hence, this study provides the first systematic evaluation of Arabic morphological analyzers on Moroccan Arabic diminutives. The dataset consists of approximately 200–300 Moroccan Arabic sentences, each containing one or more diminutive forms, manually annotated for lemma, part of speech (POS), and gloss. The analyzers under evaluation include MADAMIRA and CAMeL Tools. Analyzer performance will be evaluated using coverage (the proportion of diminutive tokens recognized), accuracy (for lemma, pos, and gloss) and precision, recall, and F1 (to measure the accuracy of morphological analyses against a gold-standard annotation). The evaluation is currently in progress, and preliminary observations suggest limited recognition of diminutive forms by existing analyzers. Full results and analysis will be presented at the conference. An error analysis will also be conducted to identify patterns of under-recognition and misanalysis. The results are expected to highlight systematic gaps in current analyzers, underscoring the need for developing morphological tools with broader dialectal coverage.

The total assimilation of the Arabic definite marker l- is triggered by stem-initial coronal consonants, exceptfor the fricative [ž] (an affricate in some dialects) which never assimilates in Standard Arabic (StA) butmay or may not assimilate in spoken varieties. This pattern is illustrated by the StA and Moroccan Arabic(MA) cognates in (1). In MA, assimilation applies in (1b, d) but fails in (1f). Freeman (2016), Harrell(1962), and Heath (1987) propose that forms such as (1f) are borrowings from StA and therefore lexicallyspecified as exceptions to the assimilation process. In a recent corpus study, however, Nirheche (2025)shows that the behavior of [ž] in MA is gradient rather than categorical: assimilation occurs in 96% of caseswhen [ž] precedes a consonant (1b), 84% before a schwa (1d), and 37% before a full vowel (1f). Heproposes that in MA, l-ž assimilation is phonologically rather than lexically conditioned.(1) Standard Arabic Moroccan Arabica. /l-žamal/ [lžamal] *[žžamal] b. /l-žməl/ [žžməl] *[lžməl] ‘camel (def.)’c. /l-žarab/ [lžarab] *[žžarab] d. /l-žərba/ [žžərba] *[lžərba] ‘scabies (def.)’e. /l-žanuub/ [lžanuub] *[žžanuub] f. /l-žanub/ *[žžanub] [lžanub] ‘south (def.)’In this paper, we depart from both accounts and propose that MA l-ž assimilation is best captured within astratified yet permeable lexicon, organized into a nativized MA ‘core’ and a StA ‘periphery’ stratum. Thisview situates MA within the broader theory of lexical stratification (Becker & Gouskova, 2016; Hsu &Jesney, 2018; Itô & Mester, 1995, 1999; Jurgec, 2010; Smith, 2018) while adapting it to a diglossicarchitecture in which strata, part of the overall grammatical system, are non-nested but crucially interactthrough porous boundaries (see Idrissi et al., 2021). This approach finds strong and hitherto undiscussedsupport in near-minimal cognates such as listed in (2): core stems systematically undergo assimilation(2a,c), while peripheral ones resist it (2b,d). Undergoing assimilation is thus determined neither by simplevariety affiliation (borrowed vs. non-borrowed) nor about its phonological environment (i.e., the makeupof the stem-initial syllable), but by lexical organization within a diglossic grammar.(2) Core items (MA) Periphery items (StA)a. /l-žəbha/ [žžəbha] (pl. žbuh) ‘forehead’ b. /l-žabha/ [lžabha] (pl. žabah-aat) ‘frontline’c. /l-žayħa/ [žžayħa] (pl. žayħ-at) ‘misfortune’ d. /l-žaaʔiħa/ [lžaaʔiħa] (pl. žawaaʔiħ) ‘pandemic’Further evidence for our permeable-stratified lexicon is found in the behavior of cognates such as in (2)across different grammatical domains. This includes: (i) semantic shift: cognates often diverge in meaning(see (2a,c) vs. (2b,d)); (ii) morphological classes: cognates may select different morphological processes(e.g., sound plural (2b,c) vs. broken plural (2a,d)); (iii) phonemic inventory: MA core vocabulary lacks theglottal stop (StA /muʔmin/ vs. MA /mumən/ ‘believer’); and (iv) syntactic distribution: periphery participlesavoid the MA circumfixal negative marker ma—š and selecting the particle maši (maši ħaaʔir ‘notconfused’ but *ma-ħaʔir-š). Notably, borrowings tend to enter the core and behave accordingly (e.g.,French jaquette > /l-žakita/ = [žžakita] ‘jacket’ and journal > /l-žurnal/ >žžurnal ‘newspaper’). Thesepatterns show that stratification in MA extends beyond phonology to morphology, semantics, and syntax.Crucially, we argue that the stratification defended in our model is not static: the core and peripheral strataremain parts of a single ‘converging’ synchronic grammar, where the MA and StA grammatical subsystemsmay be distinct yet, because their borders are fluid, interact dynamically at all levels of structure andprocessing (Idrissi et al., 2021). Categorical patterns of the type in (1) are characteristic of lexical itemsunambiguously housed either in the core or the periphery. The appeal of the proposed model, however, liesin the fact that it also predicts cases where lexical items, be they native or borrowed, can showdual/ambiguous behavior whereby they can easily shift between core and periphery depending onextragrammatical factors such as speaker literacy level, register or context, language attitudes, age, gender,or even subdialect (e.g., Arabic /l-žil/ ‘generation’ can be lžil or žžil; and French /l-žinirik/ ‘credits’ can belžinirik or žžinirik), reflecting partial interaction between the two systems of the speaker’s mental grammar.

Arabic dialects exhibit remarkable variation in how they retain lexical items from Standard Arabic (SA). While some regional dialects preserve SA lexemes with their original meanings, others demonstrate significant semantic shifts, complete replacement, or specialized usage (Versteegh, 2014; Holes, 2004). This study presents a systematic investigation of differential semantic retention patterns across major Arabic dialects, examining how and why identical SA lexemes evolve differently across linguistic communities.We introduce a comprehensive, multi-dialectal dataset that documents semantic retention patterns for 2,283 SA lexical items across 19 regional dialects, including Egyptian, Tunisian, Gulf, Levantine, Moroccan, and Yemeni varieties (Brustad, 2000). For each lexeme, we document: (1) the SA meaning, (2) dialectal meanings, and (3) frequency of usage. We leverage the data provided by theonline dialectal dictionary mo3jam.com, which contains words and expressions from 19 regionaldialects. Each linguistic item has received multiple votes from users, indicating whether they agree or disagree with its inclusion in the respective dialect. This dialect verification process is further vetted in two additional ways: [1] against multiple standard datasets for Arabic dialects, and [2] eliciting judgements from native speakers of each dialect.Our comparative analysis reveals three primary retention patterns: (1) Shared Retention – lexemes maintaining identical semantic content across dialects (e.g., akala "to eat"); (2) Differential Shift – lexemes exhibiting distinct semantic evolution in different regions (e.g., bakhīl: "stingy" in Egyptian vs. "lazy" in Tunisian) (Traugott & Dasher, 2001); (3) Synonymous Retention - different dialects selecting a lexeme from SA near-synonyms to express a given concept (e.g., ‘sleep’ for example is ‘yana:m’, ‘yarqudh’ and ‘yan3as’ in Egyptian, Tunisian and Moroccan dialets, respectivly; and (4) Selective Retention – SA lexemes with multiple meanings where different dialects retain only one of those meanings (e.g., shaḥīḥ: "scarce" in Egyptian vs. "stingy" in Tunisian) (Nerlich & Clarke, 2003; Geeraerts, 2010).This research contributes to Arabic dialectology by providing: (1) the first systematic cross-dialectal semantic retention dataset; (2) evidence-based patterns of semantic divergence; and (3) a framework for understanding why certain lexical items remain stable while others undergo rapid semantic change (Owens, 2013). The dataset is designed to support future research in comparative dialectal studies, sociolinguistic variation, and Arabic natural language processing (Habash, 2010).

Khuzestani Arabic (KhA) is a gΩlΩt dialect of South Mesopotamian Arabic spoken by Iranian Arabs in thesouthwestern province of Khuzestan. Despite its typological importance, KhA has remained largelyunderstudied due to restricted academic access and sociopolitical marginalization (Leitner, 2022; Matras &Sakel, 2007). This dialect employs several light verbs (e.g., gɑʕΩd ‘standing’, gɑm ‘stood’, rɑħ ‘went’, dˤæɫ‘stayed’, and ɣΩdæh ‘outpaced’) that denote various aspectual meanings (e.g., inceptive, durative) and thattake lexical verbs as complements which denote the event itself (1).(1) gɑm yԥ-rkԥdˤstood.3sg.masc.pfv impf-run.3sg‘He started to run.’Light verbs may optionally host second-person dative clitics læ-k / lΩ-tʃ (singular masculine and feminine,respectively) or l-kæm / l-tʃæn (plural masculine and feminine, respectively) (2), which may also be hostedby the main verb (3). Attaching the clitic to the light verb is the unmarked pattern.(2) gɑm-læ-k yԥ-rkԥdˤ (3) gɑm yԥ-rkԥdˤ-læ-kstood.pfv-for-2sg.masc impf-run.3sg‘He started to run.’I argue that these clitics (i) are not genuine benefactive or dative markers, but (ii) serve a pragmatic-discourse function, namely, addressee involvement and event relevance. Four lines of evidence support thisclaim. First, the optionality of these clitics and their free occurrence with transitive, intransitive, andditransitive verbs indicate that they are not selected arguments. Second, whether they attach to the lightverb or to the main verb does not affect the truth conditions of the clause, leaving its core meaning largelyunaffected. If they were genuine benefactives, their attachment site would necessarily correspond to theverb whose semantics they modify. Instead, their positional flexibility indicates a pragmatic or discourse-driven function. Nevertheless, a slight semantic shift is observed: attachment to the light verb emphasizesaspectual information such as event initiation or actualization, while attachment to the main verbemphasizes the event itself. Third, unlike in other Arabic dialects, KhA clitics are restricted to second-person forms, to the exclusion of first and third-person forms, reflecting an addressee-oriented pragmaticfunction centered on addressee involvement. The infelicity of (3), which bears a non-second-person clitic,thus follows naturally from this restriction:(3) * gɑm-ԥl-hæh yԥ-rkԥdˤstood.3sg.pfv-for-3sg.fem impf.run.3sgFinal supporting evidence comes from negation: when these clitics co-occur with negated predicates, theresult is semantically infelicitous (4):(4) # mɑ-gɑm-læ-k yԥ-rkԥdˤneg-stood.pfv-for-2sg.masc impf.run.3sg‘He didn’t start to run.’I argue that this infelicity arises because these clitics presuppose the initiation or actualization of the event,which conflicts with sentential negation asserting non-realization. This clash, thus, results in pragmaticpresupposition failure. Negation becomes felicitous only when it contrasts with a presupposed affirmative,as in Givón’s (2001, p. 370) example, where negation conveys a contrary-to-expectation inference:(5) A: What’s new?B: My wife isn’t pregnant.Negation thus presupposes that the corresponding affirmative proposition (“My wife is pregnant”) wascontextually expected. Similarly, in KhA, second-person form clitic construction presupposes an initiatedor real event; when negation denies that event, the presupposition fails, yielding pragmatic anomaly.KhA contributes new empirical evidence to the growing cross-dialectal literature on clitics (e.g., Bar Moshe,2021; Haddad 2020a; Haddad 2020b), showing that KhA dative clitics serve as addressee involvement andevent salience markers rather than argument structure. These findings broaden our understanding of thesyntax-pragmatics interface in Arabic through discourse-oriented morphology.

Indexicals in the complements of attitude verbs can shift their reference from the utterance context to the reported speech context in many languages (e.g. Amharic, Schlenker 1999; Zazaki, Anand and Nevins 2004; Anand 2006; Nez Perce, Deal 2014, 2019; Korean, Park 2016). The goal of this paper is to establish that d:a:ri:dZa (Moroccan Arabic, MA) also exhibits indexical shift in complement clauses, as shown in (1) with time and locative indexical adverbs —a phenomenon to our knowledge not reported before. (1) a. On Monday, June 3rd, Sana says: “I will leave tomorrow.” On June 10th, I tell Ayoub:ttninmondaylli:Cfa:tpassedsanasanaga:l-tsay.PAST-F3Sli-jja:to-meb@lli:C°Ga-tmSi:FUT-F3S.walkGdda:tomorrow‘Last Monday, Sana told me that she would leave the next day.’ b. In Mirleft, Ayoub says: “people are nice here”. In Nantes I utter:m@lli:whenka:nbe.PAST.M3Sf-mi:rl@ftin-mirleftga:lsay.PAST.M3Sb@lli:C°nnaspeoplemzja:n-innice-PLHna:ja:here‘When he was in Mirleft, Ayoub said that people are nice there.’ In (1a), the indexical time adverb Gdda: ‘tomorrow’ refers to the day after the day of the reported speech event (June 4th), not the day following the day of speech (June 10th). In (1b), Hnaja: ‘here’ designates the location of the attitude holder (Mirleft), not that of the speech act (Nantes). We offer 6 arguments to establish that this perspective shift has the hallmark properties of IS, and is not reducible to alternative analyses –in particular, (partial) quotational analyses (i.e. (iii-vi)):i. It is restricted to complements of speech report verbs (’say’, ’tell’) and thus unavailable in other subordinate contexts.ii. Shifted adverbs are true indexicals: they cannot be analyzed as containing a bindable anaphor responsible for their shifted/anaphoric reading, since they cannot be bound by/covary with a quantifier, unlike t@mma ‘there’ or lG@dliH ‘the next day’ (cf. Kaplan 1989).iii. Material in embedded clauses appears in nonverbatim form, even in clauses containing a shifted indexical (Shklovsky and Sudo 2014).iv. Shifted indexicals are possible in embedded environments where an argument is interpretedde re (Deal 2019).v. Availability of long-distance extraction from an embedded clause containing a shifted indexical (cf. Sudo 2012), as shown in (2).vi. Shifted indexicals obey the Shift-Together Constraint (Anand and Nevins 2004) which requires all indexicals in the same embedded domain to shift together: both of the embedded adverbs in (2) can shift together ((a)), remain unshifted ((b)), but mixed readings ((c-d)) are ungrammatical. Crucially, the shift cannot be due to direct quotation, as the wh-phrase has been moved long-distance out of the embedded clause and quotations are islands for movement (Sudo 2012). (2) Shifted context: Speaker asks this question on a Saturday in Nantes.Sku:nWhoga:l-tsay.PAST-F3Ssanasanattninmondaylli:Cfa:tpassedf-mi:rl@ftin-mirleftb@lli:C°Ga-jSu:f-u:FUT-see.3SM-himajju:bayoubHna:ja:hereGdda:?tomorrowLit: ‘Who did she say, last Monday, that Ayoub would see here, tomorrow?’ a.‘here’ = Nantes, ‘tomorrow’ = Sunday *c.‘here’ = Mirleft, ‘tomorrow’ = Sunday b.‘here’ = Mirleft, ‘tomorrow’ = Tuesday *d.‘here’ = Nantes, ‘tomorrow’ = Tuesday

There is a well-known distinction between two answering systems, often referred to as the truth-based system and thepolarity-based system (Jones, 1999; Holmberg, 2016). These systems differ in how they answer yes-no questions like (1).(1) Q: Does John not drink coffee?In truth-based languages like Japanese, the negative proposition (John doesn’t drink coffee) is confirmed by ‘yes’ and thepositive one (John drinks coffee) by ‘no’. In polarity-based languages like Swedish, ‘no’ confirms the negative proposition,and a special affirmative particle ‘yes.REV’ confirms the positive proposition. Holmberg (2016) links this contrast to theposition of negation (i.e., NEG is low inside vP in truth-based languages and high outside vP in polarity-based ones). Basedon novel data from Hijazi Arabic, we propose that answer particles target the semantically salient proposition, not negation.In truth-based languages, this is the entire proposition (including negation), whereas in polarity-based languages, it is thepositive proposition even when the question is negative, as illustrated in (1'). Following Krifka (2013), we treat answerparticles as operating on propositions introduced by the question, but show that in Hijazi Arabic, salience is fixed as thepositive alternative, independent of discourse.(1') a. Truth-based: Salient proposition is John doesn’t drink coffeeA1: ‘yes’ (= he doesn’t drink coffee) A2: ‘no’ (= he drinks coffee)b. Polarity-based: Salient proposition is John drinks coffeeA1: ‘yes’ / ‘yes.REV’ (= he drinks coffee) A2: ‘no’ (= he doesn’t drink coffee)Hijazi Arabic, like many polarity-based languages (Holmberg 2016), has a three-particle answering system: ʔiwa ‘yes’confirms the positive proposition, laʔ ‘no’ the negative, and ʔilla ‘yes.REV’ reverses the negative polarity to confirm thepositive proposition. While ʔiwa and laʔ are felicitous with neutral questions (2), ʔilla is licensed only under negation (3).(2) Q: saːg xaːlid ʔas-sajjaːra?drove.3SM Khalid the-car‘Did Khalid drive the car?’A1: ʔiwa. A2: laʔ. A3: #ʔilla.yes no yes.REV‘Yes.’ (= he drove the car) ‘No.’ (= he didn’t drive the car) ‘Yes, he did.’(3) Q1: xaːlid kaːn maː ji-ʒri fi-l-leːl? Q2: xaːlid maː kaːn ji-ʒri fi-l-leːl?Khalid was.3SM NEG 3SM-run at-the-night Khalid NEG was.3SM 3SM-run at-the-night‘Was Khalid not running at night?’ ‘Was Khalid not running at night?’A1: #ʔiwa. A2: laʔ. A3: ʔilla.yes no yes.REV‘Yes.’ ‘No.’ (= he wasn’t running at night) ‘Yes, he was.’The questions in (3) also show that negation in Hijazi Arabic may occur inside vP (3Q1) or outside it (3Q2). In both cases,only laʔ and ʔilla are felicitous, while ʔiwa is not, indicating that the syntax-based account cannot explain the distributionof answer particles. Following Krifka (2013), we analyse answer particles as propositional anaphors referring to discoursereferents introduced by the question. In neutral questions, LF introduces a positive proposition d, which ʔiwa assertsand laʔ rejects (¬d). In negative questions, LF introduces d and its negative counterpart d′; ʔilla asserts d and laʔ asserts ¬d.The observation that ʔilla is licensed only under negation indicates that d′ is introduced in such contexts, while the infelicityof ʔiwa suggests it is blocked when d′ is present. Crucially, this does not make d′ the salient alternative. Since ʔilla stillasserts d, the positive proposition remains semantically salient. If d′ were salient, we would expect the particle system toshift accordingly: ʔiwa should then be able to assert d′ (= ‘Yes, Khaled wasn’t running at night’), and laʔ should be able toreject it (= ‘No, he was’). Discourse-biased contexts like (4) further support this analysis: regardless of bias, ʔilla asserts d,laʔ asserts ¬d, and ʔiwa remains infelicitous. Thus, in Hijazi Arabic (and perhaps in other polarity-based languages), theanswering system is not sensitive to discourse-level salience, contrary to what we would expect in truth-based languages.(4) Context: The speaker knows the hearer drinks coffee every morning, yet the cup remains untouched at noon.Q: maː ʃəribt gahwa lissa?NEG drank.3SM coffee yet‘Did you not drink coffee yet?’A1: laʔ. ‘No.’ (= I didn’t drink coffee) A2: ʔilla. ‘Yes, I did’ A3: #ʔiwa. ‘Yes.’These observations suggest that answer particles are determined not by structure (Holmberg, 2016) or discourse (Krifka,2013), but by how languages encode propositional salience. The Hijazi Arabic pattern thus shows that cross-linguisticvariation in answer particles reflects differences in which proposition the grammar selects as semantically salient.